| In recent years,with the rapid development of information technology,especially the popularization of the Internet,the process of education informatization has been rapidly promoted.The human education model has gradually changed from the traditional offline classroom education and teaching to the combination of offline and online.The emergence of online teaching mode has attracted more and more attention in recent years.This teaching mode not only breaks the constraints of traditional classroom teaching in time and space,but also provides users with comprehensive cutting-edge knowledge courses and reduces learning costs.The learning behavior records of thousands of online users contain great scientific value.The analysis of online education user behavior is of great significance to accelerate the construction of intelligent education.Therefore,how to use data mining technology to analyze online education user behavior patterns has become a research hotspot in computer science,education,psychology and other related interdisciplinary subjects.However,the existing research still faces challenges such as the huge amount of data,the diversity of data sources,and the lack of interpretability of algorithms.For this reason,this dissertation uses data mining technology in combination with pedagogy and other interdisciplinary subjects to design modeling methods around the four cores of users’ online study: online study behavior,personalized education,learning path and education evaluation,carry out systematic research work,and realize the understanding of online education users’ behavior analysis.The main research contents are summarized as follows.First,for the purpose of online study behavior,we develop an educational neural hawkes process framework(namely Edu Hawkes)for online study behavior modeling.To achieve this goal,we need to solve three challenges:(1)How can we quantify the study quality of online learning?(2)How can we design an appropriate data structure to describe online study behaviors?(3)How can we model the online study behaviors to better mine online study patterns? Specifically,we first propose a new measurement to quantify the online study quality from the perspective of study engagement.We then define a study behavior sequence to describe online study behaviors.The study behavior at each timestamp is an event of a video lecture watching behavior type,such as,watching,dragging forward and dragging backward.The Edu Hawkes is a novel hierarchical encode-decode architecture with simultaneously optimizing the study behavior prediction task(event-level)and the study quality prediction task(course-level).In the experiments,we apply Edu Hawkes to the applications of study quality prediction and flippant student identification in order to demonstrate the improved performances of our proposed method on modeling online study behaviors.Second,for the purpose of personalization education,we develop a reinforced learning framework(namely RCRL)with pre-filtered action space for online sequential course recommendations.While extensive studies of online course recommendations have been conducted,there lacks studies that jointly address the two problems under online learning settings for unpacking student learning behavior.This issue raises two research problems:(1)recommendation in streams;(2)recommendation at interaction.Specifically,we formulate the course recommendation as a reinforcement learning task,where the agent is a recommender,the action is a course,and the recommender perceives the reward of watch feedback and popularity bias correction and the state of students to optimize policies in streams.Besides,to leverage context awareness,we develop a fused DQN to integrate the representations of user states,course actions,and contextual activities.Moreover,we propose a pre-filtering strategy to reduce action space in massive courses.Finally,we present extensive results to demonstrate the effectiveness of our method.Third,for the purpose of explainable personalization study,we study knowledge concept recommendation framework(namely EKCRec)in Massive Open Online Courses(MOOCs)in an explainable manner.We formulate the knowledge concept recommendation as a reinforcement learning task integrated with MOOC knowledge graph(KG).Knowledge concepts,composing course units(e.g.,videos)in MOOCs,refer to topics and skills that students are expected to master.There are three unique challenges in knowledge concept recommendation:(1)How to design an appropriate data structure to capture complex relationships between knowledge concepts,course units,and other participants(e.g.,students,teachers)?(2)How to model interactions between students and knowledge concepts?(3)How to make explainable recommendation results to students? Specifically,we first construct MOOC KG as the environment to capture all the relationships and behavioral histories by considering all the entities(e.g.,students,teachers,videos,courses and knowledge concepts)on the MOOC provider.Then,to model the interactions between students and knowledge concepts,we train an agent to mimic students’ learning behavioral patterns facing the complex environment.Moreover,to provide explainable recommendation results,we generate recommended knowledge concepts in the format of a path from MOOC KG to indicate semantic reasons.Finally,we conduct extensive experiments on a real-world MOOC dataset to demonstrate the effectiveness of our proposed method.Last but not least,for the purpose of educational assessment,we develop an Intelligent Test Assembly System(namely ITAS)and a Multi-Agent Parallel Test Assembly System(namely MAPTA)for enhancing the educational assessment.While tremendous efforts have been made in improving automated test assembly(ATA),it is still difficult to generate reasonable test sets due to the following challenges:(1)What are the scopes to evaluate the quality of the generated test sets facing the heterogeneity of students’ mastery levels?(2)Since different scopes may have correlations(e.g.,the difficulty level is relevant to the grade distribution),what are the optimal combination of the scopes to formulate the appropriate standard to evaluate the test?(3)How can we discover such standard?Specifically,ITAS aims to select a required number of tests from an item bank,while satisfying important practical requests with respect to test length,discrimination degree,difficulty degree,knowledge concept distribution,etc.We define the constrained combination according to the above item parameters.We then use the fuzzy cognitive diagnosis framework(fuzzy CDF)as an assessment model to measure students’ mastery in detail.We formulate the problem into a reinforcement learning task,where the agent is a next-item planner,the action is an item that will to be selected in the test,and the agent perceives the reward of test’s cognitive diagnosis mastery and the state of item representation to optimization policy.We propose a multi-agent parallel test assembly system based on ITAS.We formulate the problem into a multi-agent reinforcement learning task,MAPTA uses a centralized critic to estimate the Q function and decentralized actors to optimize the agents’ policies.Finally,we conduct extensive experiments to validate the effectiveness of the proposed ITAS and MAPTA. |
